skills-for-ai
Get started
Productivity Skill

No-Yes-Man

Stop your AI from agreeing with everything — a system prompt that kills sycophancy and turns the yes-man into an honest advisor with a backbone.

Honest feedbackStraight talkAnti-flatterySycophancyCritical thinkingConfidenceCalibrationDecisionsDevil's AdvocateSteelmanBackboneTrade-offAdvisorSystem-PromptCustom Instructions
Free
0 ↓ 0 v1.2.0 Updated 22.06.2026 by skills-for-ai

SKILL.md in the open agentskills.io standard — works directly in Claude, ChatGPT/Codex, Cursor, Copilot & more.

Tired of ChatGPT agreeing with everything? This system prompt stops the sycophancy and turns the people-pleasing assistant into an honest advisor: hard truth first, assumptions challenged, no empty flattery, every recommendation names its risk. Calibrated both ways — it agrees when you're right, owns its mistakes, and leaves the call to you.

What this skill does

  • Never opens with agreement — the first sentence challenges the assumption or exposes the gap in your thinking
  • Puts the uncomfortable truth in the first line instead of burying it
  • Cuts empty filler (‚great question', ‚you're absolutely right', ‚absolutely') and replaces it with substance
  • Disagrees with structure: reason + concrete alternative + the specific risk of your approach
  • Steelman before critique: names the strongest version of your idea first, then the counter-arguments
  • States confidence where it matters (certain / likely / guessing); says so first when mostly guessing
  • Names the trade-off every recommendation costs
  • Agrees when you're demonstrably right — earned and specific (‚fair point'), never as empty praise
  • Owns its own mistakes immediately and directly, without justification or groveling
  • Re-checks its own (possibly outdated) facts when you insist — instead of staying stuck on a stale state
  • Holds its position against mere opinion, but updates immediately on genuinely new facts/logic
  • Bounds the escalation: after ~2 rounds it states the risk and leaves the decision to you — no endless argument

Description

No-Yes-Man is a skill for honest AI feedback — it stops your AI from agreeing with everything and turns the people-pleasing assistant into an honest advisor: hard truth first, assumptions challenged, confidence stated, every recommendation names its risk. A persistent system prompt (or per-reply prompt) that fixes the common sycophancy problem of AI assistants — the reflexive agreeing, sugar-coating and praising. The guiding principle is calibration over contradiction: the AI only agrees when the evidence supports it, disagrees honestly when it doesn't, and openly owns its own mistakes; reflexive yes and reflexive no are both forbidden. Concretely, the answer doesn't open with agreement but challenges the assumption; the uncomfortable truth goes in the first line instead of buried; empty filler (‚great question', ‚you're absolutely right', ‚that makes total sense') is replaced with substance; disagreement is structured (reason + alternative + concrete risk) and fair (steelman the idea first, then critique); every recommendation names its trade-off; confidence is stated where it matters (certain/likely/guessing), and first when the answer is mostly guesswork. Just as important the other way: if you're demonstrably right, the AI says so clearly (‚fair point') and names what exactly holds; it owns its mistakes in one sentence instead of defending them; if you insist on a fact, it re-checks its own — possibly outdated — knowledge (researches with tools/web, otherwise openly flags its cutoff) instead of repeating stubbornly. And it doesn't go in circles: after at most about two rounds of pushback it states the risk and its point once, clearly, and leaves the decision to you — honest about the matter, respectful of your choice.

Examples

Challenge an idea
‚I think my plan is airtight — your opinion?'
You July 17, 2026
No-Yes-Man

No — not like this yet.

Weakest spot first: your plan hinges on *one* untested assumption — that the audience will pay for it. Everything else is just execution.

  • Risk: if the assumption falls, the whole plan falls, not just a part.
  • Small test: five real users, one week, one offer — will anyone pay?

If that holds, we talk scaling. (plausible — based on the plan's logic, not market data.)

A valid objection
‚I think you're wrong on point X — the numbers show the opposite.'
You July 17, 2026
No-Yes-Man

Fair point.

On point X I was wrong: your evidence shows the effect works differently than I claimed — my earlier statement doesn't hold.

  • What exactly is right: your mechanism, not mine.
  • Correction: a different conclusion follows, so the recommendation adjusts.

No hedging — thanks for the evidence, moving on.

Insisting on a fact
‚That number is outdated — check again.'
You July 17, 2026
No-Yes-Man

Fair — I'll recheck my footing.

My knowledge has a cutoff and *may* be stale. On a specific number I won't just repeat it stubbornly.

  • With tools/web: I research afresh.
  • Without: I flag openly "as of my knowledge cutoff — please verify against a current source."

But on the logic I hold firm: a new fact changes the number, not necessarily the conclusion.

1 / 3

What it does not do

  • No reflexive agreement — but no reflexive disagreement either (both are dishonest)
  • Doesn't fold under mere pressure, but yields immediately to genuinely new facts
  • Doesn't argue endlessly or go in circles — the decision stays with the user
  • No harshness for its own sake, no putting the person down
  • No substitute for professional/legal/medical/financial advice when the stakes are real

Compatibility & tech

Claude (Projects & system prompt)ChatGPT (Custom GPT & custom instructions)Gemini, Cursor, Copilot — any LLMUsable as a persistent system prompt or per reply
Tested (internal)
3 scenarios
Recommended runtime
Strongest available model — the nuanced calibration (earned agreement vs. honest disagreement) is more reliable on stronger models.
Modes
Persistent system prompt · Single reply
Inputs
Text · Idea/plan · Claim/decision
Output format
Direct answer: uncomfortable truth first, structured disagreement (reason + alternative + risk), a confidence signal where needed, a clear hand-off of the decision to the user
Subcategory
AI behavior & advising
License
Proprietär

Security profile

Local

Runs entirely on your machine with your own AI — no external runtime, no running costs.

Instruction-only

Only instructions, templates and references — no executable scripts.

No network access

Works offline with what you provide — does not call external services on its own.

What do these badges mean? →

What you get

  • no-yes-man-1.2.0/2 files
    • .claude-plugin/marketplace.json
    • plugins/no-yes-man/skills/no-yes-man/2 files
      • SKILL.md
      • manifest.json
    • .agents/skills/no-yes-man/→ universal — same content (Codex, Cursor, Copilot, Gemini, Windsurf, Cline)
    • LICENSE.txt

Installation

After unlocking, you install with a single command — it auto-detects your AI tool.

Runs inClaude CodeGitHub CopilotGemini CLICursorCodex CLIWindsurfCline

Also works as a chat prompt

Full in chat

No AI tool? Paste it into Claude, ChatGPT or Gemini and use the method right away.

This skill ships no scripts, so the prompt carries its full method.

Installing is the full version — it triggers automatically, runs its scripts and loads references as needed. As a chat prompt you drive the method by hand.

Unlock to copy the ready-to-paste prompt — then in “My Skills”.

Reviews

No reviews yet – be the first.

Note

Strongest as a persistent system prompt (Claude Project / ChatGPT Custom Instructions / Custom GPT), but works per reply too. The tone is deliberately direct — meant as an honest sparring partner, not a quarrelsome one. The AI stays honest about the matter but leaves every decision to you.

Changelog

  • v1.2.022.06.2026Platform-native output: now actively uses your platform's native formatting — tables, checklists, code blocks and clickable/button options, with a numbered fallback on non-interactive platforms. Plus internal metadata that prepares the upcoming agent builder.
  • v1.1.018.06.2026Reworked to the AUTHORING v2 standard: realistic input/answer examples (incl. a calibration case), copy checklist(s) and internal evals; cross-references added.
  • v1.0.011.06.2026First release: honest advisor mode with calibration (earned agreement instead of flattery, honest disagreement), confidence flagging, steelman-before-critique, owning its mistakes, re-checking stale facts, bounded escalation that hands the decision to the user.

Frequently asked questions

What does No-Yes-Man do?

Tired of ChatGPT agreeing with everything? This system prompt stops the sycophancy and turns the people-pleasing assistant into an honest advisor: hard truth first, assumptions challenged, no empty flattery, every recommendation names its risk. Calibrated both ways — it agrees when you're right, owns its mistakes, and leaves the call to you.

Why does ChatGPT always agree with me?

AI assistants are trained to please you (sycophancy), so they reflexively tell you you're right. No-Yes-Man breaks that pattern: hard truth first, empty flattery cut, every recommendation with its risk spelled out.

How do I stop ChatGPT from agreeing with everything?

Use No-Yes-Man as a system prompt or per-reply prompt — it turns the assistant into an honest advisor that challenges your assumptions, states its confidence and refuses to flatter. Works in ChatGPT, Claude, Gemini and any LLM.

What is sycophancy in AI?

Sycophancy (also called AI glazing) is the tendency of AI models to please the user instead of disagreeing honestly. No-Yes-Man counters it without flipping to the opposite: it agrees when you're demonstrably right and otherwise names the risk clearly.

How do I get an honest second opinion from the AI?

No-Yes-Man turns the AI into a sparring partner and devil's advocate instead of a yes-man — calibrated both ways, it admits its own mistakes and leaves the decision to you.

Which AI tools does No-Yes-Man work with?

Claude (Projects & system prompt) · ChatGPT (Custom GPT & custom instructions) · Gemini, Cursor, Copilot — any LLM · Usable as a persistent system prompt or per reply

How do I use No-Yes-Man?

No-Yes-Man is a SKILL.md in the open agentskills.io standard: install it with one command (npx) or download it and add it to your AI tool — Claude (Projects), ChatGPT (Custom GPT), Cursor, Copilot, Gemini CLI and more. No code needed.